Sentiment-Analysis-on-Twitter-Data

Twitter Sentiment Analysis (NLP) The project follows an end-to-end ML workflow similar to a production-style pipeline.

Overview

This project performs sentiment analysis on Twitter data using Natural Language Processing (NLP) and Machine Learning techniques.
The goal is to classify tweets into positive, negative, or neutral categories.

Dataset

Source: Kaggle – Twitter US Airline Sentiment
Size: 14,640 tweets

Each tweet includes:

• Text content
• Sentiment label (positive / negative / neutral)
• Airline information
• Confidence scores

Dataset is stored in:

---

Project Structure

twitter-sentiment-analysis/ │ ├── data/ │ └── raw/ │ └── Tweets.csv │ ├── notebooks/ │ └── 01_twitter_sentiment_eda_modeling.ipynb │ ├── model/ │ ├── sentiment_model.pkl │ └── tfidf_vectorizer.pkl │ ├── README.md └── requirements.txt

---

Methodology

1. Exploratory Data Analysis (EDA)

• Sentiment distribution analysis
• Tweet length analysis
• Class imbalance inspection

2. Text Preprocessing

• Lowercasing
• Removing URLs, mentions, punctuation
• Stopword removal
• Lemmatization

3. Feature Engineering

• TF-IDF Vectorization

4. Modeling

• Logistic Regression
• Linear Support Vector Classifier (LinearSVC)

5. Evaluation

• Precision, Recall, F1-score
• Comparison of multiple models

---

Results

• Achieved ~77% accuracy
• LinearSVC showed strong performance on negative sentiment detection
• Clean and reproducible pipeline

---

Tools & Libraries

• Python
• Pandas, NumPy
• Scikit-learn
• NLTK
• Matplotlib, Seaborn

---

How to Run

pip install -r requirements.txt
## Author
Ivan Lucas